A supply-chain KPI is useful only when its number can be traced to operational evidence and an authorized owner knows what state requires action.
The dashboard is not the source of truth. Orders, promises, warehouse movements, shipment milestones, receipts, invoices, and ledger records are. The product must preserve their identity, time, corrections, and scope before assigning a status colour.
A Supply-Chain KPI Needs a Metric Contract Before It Is Allowed a Status Colour
For each KPI, write:
- decision, audience, and authorized scope;
- grain: order, line, package, shipment, SKU-location, receipt, supplier, lane, or period;
- numerator, denominator, unit, currency, and aggregation;
- event time, promise window, timezone, calendar, and observation close;
- inclusions, exclusions, partials, duplicates, corrections, and unknown states;
- source priority, raw evidence, model version, and freshness;
- target or control rule with effective dates;
- owner, permitted response, runbook, confirmation, and escalation.
Borrowed benchmarks cannot replace this contract. A red threshold is meaningful only when it represents an agreed action state for this operating context.
Order Fill Rate Means Three Different Things, So Define the Grain Before the Target
Order fill rate can mean units shipped divided by units ordered, complete lines divided by requested lines, or orders fulfilled without a backorder. These are not interchangeable.
Define how the metric treats split shipments, substitutions, partial fulfillment, customer-requested changes, cancellations, preorders, backorders, duplicate orders, and orders still inside their promise window. Preserve the original and revised promise with actor and reason.
On-time delivery also requires an accepted terminal event and a governed promise. Carrier “delivered,” proof of delivery, customer acceptance, and warehouse receipt may represent different milestones. Define early delivery, appointment windows, weekends, local time, grace periods, and missing scans.
A perfect-order metric combines several conditions at once. Publish the components and failed condition so an owner can distinguish incomplete fulfillment, lateness, damage, and documentation error. Do not call it a gold standard without defining the exact contract.
The logistics metrics guide should preserve these order, shipment, and milestone identities through drill-through.
Inventory Is a Ledger of Movements, Not the Last Quantity a Warehouse System Reported
Inventory is not merely the last quantity received from a warehouse system. Model receipts, picks, packs, shipments, returns, transfers, adjustments, reservations, quarantine, damage, cycle counts, purchase orders, and in-transit states.
Available, on-hand, allocated, reserved, safety stock, and available-to-promise need separate definitions. Retain movement ID, source event time, received time, SKU, unit, location, quantity, reason, acceptance state, and correction link.
Inventory turnover and days of inventory require an agreed stock basis, cost basis, time window, seasonality treatment, and behavior for zero or negative demand. A fixed “healthy” number is not universal across perishable goods, spare parts, fashion, or critical components.
Stockout risk is a prediction when it depends on future demand and lead time. Show model version, horizon, input freshness, confidence or scenario, and fallback. Keep observed stock states distinct from predicted risk.
Supplier On-Time Delivery Needs an Accepted Receipt Event, Not an Average Lead Time
Supplier on-time delivery needs a purchase-order promise, revision history, accepted receipt event, timezone, quantity tolerance, quality acceptance, and rules for partial delivery. Separate supplier-caused delay from buyer changes, customs holds, carrier failures, and disputed receipts where the decision requires it.
Lead-time variance must name its start and end milestones: order creation, supplier acceptance, dispatch, port departure, arrival, warehouse receipt, or quality release. Average lead time without its distribution and observation population can hide operational risk.
Segment by supplier, item class, origin-destination lane, transport mode, season, and service level only where sample size and authorization allow. Show missing and unmatched events instead of silently excluding them.
For broader operational patterns, use the supply-chain dashboard guide.
Cost Per Order and Cash-to-Cash Come From Accounting Close, Not From Live Events
Cost per order may include warehouse labor, packaging, storage, handling, transport, fuel surcharge, duty, returns, damage, software, or allocated overhead. Define eligible costs, allocation basis, currency source, conversion time, denominator, and accounting close.
Cash-to-cash cycle time combines inventory, receivables, and payables measures. Its inputs come from accounting policy and period close, not necessarily live operational events. Preserve entity, calendar, consolidation, intercompany, and restatement rules.
Use the appropriate dashboard types: an exception queue may update by event, an operating view by decision window, and an executive cost view after reconciliation.
A Supply-Chain Alert Is Not a Resolution Until It Names an Owner and a Runbook
An alert is not resolution. Every actionable state should record affected object, evidence, severity, owner, permitted response, runbook, review time, and expected confirmation.
Deduplicate repeated signals, suppress known maintenance where justified, and preserve acknowledgement, assignment, action, retry, handoff, and closure history. When the expected source or milestone does not change, keep the exception open or escalate it.
Show partial, stale, unavailable, revised, and recovered states. A green tile produced from stale or incomplete sources is not healthy.
The real-time dashboard guide should be applied only after the event and action contract is clear.
Carriers, Suppliers, and Consignees Share Records, So Scope Has to Be Tested Per Role
Customer-facing supply-chain analytics may serve shippers, carriers, suppliers, warehouses, consignees, and internal operators from overlapping records. Host identity must map to permitted organization, role, facility, order, shipment, fields, metrics, and actions in trusted services.
Test isolation across filters, direct identifiers, drill-through, cache, API, shared links, saved views, CSV and PDF exports, scheduled recipients, alerts, errors, logs, and support tooling. A carrier may see a shipment without seeing commercial cost; a consignee may see ETA without supplier performance; one customer must never infer another's volume from counts or filter options.
Adjacent manufacturing operations need the same event, metric, scope, and exception discipline.
When the ERP, the Warehouse System and the Carrier Disagree, Pick One and Say Which
Three systems will report the same shipment differently, and all three will be internally consistent. The ERP knows what was promised, the warehouse system knows what left the building, and the carrier knows what it scanned. None of them is wrong; they are answering different questions with the same noun.
A dashboard that averages them produces a number no one can reproduce. Name a system of record per metric instead, state it beside the number, and keep the others available as a reconciliation view rather than as inputs to a blended figure. When someone challenges a number, the question then becomes "does the record match" rather than "which of our four numbers is real", and that is a question with an answer.
Four Tiles Someone Acts On Beat Twenty Tiles Someone Scrolls Past
The board that gets abandoned is rarely the one missing a metric. It is the one where nothing on screen maps to a decision anybody makes this week.
Choose the first tiles by working backwards from the actions available: what would make someone call a supplier, expedite a shipment, hold a release, or reallocate stock. A metric that cannot change one of those is reference material and belongs a click deeper, not on the front screen. Adding it later is easy; removing it after people have learned to ignore the board is not.
Prove the Dashboard on One Real Order Flow, Corrections and All
Use one representative order flow with known promises, movements, shipment events, receipts, costs, and corrections:
- Reconcile source totals and unmapped orders, SKUs, locations, milestones, and suppliers.
- Calculate expected metrics at order, line, shipment, SKU-location, and period grain.
- Authenticate an allowed role and deny another customer, partner, field, and action.
- Complete the intended exception, comparison, drill-through, export, or handoff.
- Exercise duplicate scans, partial shipments, revised promises, late receipts, missing events, source outage, stale cache, retry, and recovery.
- Test realistic data volume, concurrency, expensive allowed filters, extreme labels, mobile layout, keyboard access, and non-visual equivalents.
- Preserve raw evidence, rule versions, query identifiers, owners, runbooks, and rollback criteria.
Measure task completion, interpretation errors, unresolved exception age, reconciliation variance, denied-access results, reliability, and confirmed operational response.
The Bottom Line: One Decision and One Real Order Flow Beat a Full Board of Tiles
A supply-chain dashboard is an evidence and ownership system, not a catalogue of green tiles. Its KPIs need explicit grain, milestones, time, exclusions, source, threshold, scope, and response.
Start with one decision and one production-shaped order flow. Expand only after meaning, denial, workload, artifacts, exception recovery, and operating ownership pass together.
Ready to launch customer-facing analytics?
Stop losing customers to competitors with better analytics. Sumboard's customer-facing analytics platform lets you launch self-service dashboards in days, not months.
